Introduce Utils.TokenUsageCollector for token usage tracking

[nallwhy]

This PR introduces LangChain.Utils.TokenUsageCollector, a utility module designed to simplify token usage tracking.

Motivation:

Currently, while the on_llm_token_usage callback provides valuable token usage information, there isn't a readily available, best-practice approach for aggregating this data, especially in scenarios involving multiple LLM interactions within a single task. This PR aims to address this by providing a reusable and easy-to-use solution.

Benefits:

• Simplified Token Tracking: Makes it easier for developers to track token usage across multiple LLM calls.
• Inspiration for Callback Usage: Offers a practical example of how to effectively utilize the on_llm_token_usage callback.

I believe this module can be a valuable addition to LangChain, providing a more user-friendly experience for managing token usage. I'm open to feedback and suggestions for further improvements.

[nallwhy]

@brainlid
The get_callback/1 function will likely error out when used with ChatBumblebee because ChatBumblebee doesn't have a model attribute in the callback data. I'm looking for suggestions on how best to handle this case.